> I've been playing with motion correction options recently and have run
> into something that confuses me. I am working with partial brain volumes
> (12 x 7.5mm slices, EPI), so when realigning via the trilinear method I
> lose some voxels from the end slices (i.e., the inferior and superior
> slices). I presume this is because the volume is being rotated such that
> data from some voxels are moved out of the plane of the slice, leaving
> behind null voxels in their place. What I find confusing is that when I
> use MEDx to perform identical transformations on the original volumes
> (based on info in the SPM translation and rotation .dat files), again via
> the trilinear algorithm, the resulting volume looks much different than
> the one obtained via realign in SPM. Am I wrong to expect that both
> programs should yield the same image when the same transformations are
> applied to the same volume?
I can't comment on this one.
>
> A related question: One way in which I tried to get around the problem of
> lost data in the extreme slices was to resample the data (thanks Russel)
> with, say, 10 times the original number of slices, each with 1/10th the
> original slice thickness (e.g., resample 12x7.5mm slices to 120x.75mm
> slices). I expected that this transformation would simply reproduce each
> slice 10 times such that I would see 10 images each of slices 1 through
> 12. However, when I tried resampling in this way, I did not get the
> expected output volumes. Instead of 10xslice1, 10xslice2, etc, I got
> 5xslice1, 10xslice2, ..., 10xslice12, 5xnull-slice. something along those
> lines. Is there a simple way to avoid data loss in end slices of partial
> volumes?
You can gain a little bit of extra data by changing spm_vol_utils.c on lines
5 and line 550 from:
#define TINY 5e-2
to something like:
#define TINY 1e-1
Note that this will require the mex files to be recompiled (see spm_MAKE.sh).
You will also need to change the value of tiny in spm_reslice.m, lines 216
,372 and 470 and also line 158 of spm_write_sn.m.
>
> Last question: When working with a partial-brain volume of thick slices
> (like 7.5 mm), there is no way to set all the parameters necessary for
> normalization. The AC is not visible due to the low resolution and the
> inferior pole is not visible because the extent of the data does not
> include that region. So I assume that we use a whole-brain hi-resolution
> image to set the normalization parameters, then apply those parameters to
> the functional image. Is the correct order then to (1) coregister the
> partial-brain functional image with the whole-brain anatomical, (2)
> normalize the anatomical, and (3) apply the resulting transformations to
> the functional? If so, where are the normalizing transformations stored?
This sounds like a good approach, but make sure that the functional images
can be adequately registered with the structurals (<Check Reg> buton).
The rigid body transformations are stored in the .mat files of the images,
whereas the estimated spatial normalisation parameters are stored in a
*_sn3d.mat file.
>
> I'm sure it's clear from reading this that I am still trying to figure out
> what I'm doing, so if there is a anything in the literature that would
> answer these questions, I'd appreciate the reference(s), as well as any
> other suggestions from your experience.
The archives of the mailing list are probably the best source of information
on things like this:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?S1=spm
The online manual can also sometimes be of help:
http://www.fil.ion.ucl.ac.uk/spm/course/manual/man.htm
Best regards,
-John
--
Dr John Ashburner.
Wellcome Department of Cognitive Neurology.
12 Queen Square, London WC1N 3BG, UK.
tel: +44 (0)20 78337491 or +44 (0)20 78373611 x4381
fax: +44 (0)20 78131420
http://www.fil.ion.ucl.ac.uk/~john
mail: [log in to unmask]
|